Calibration of Microscopic Traffic Simulation in an Urban Environment Using GPS-Data

Stang, Christopher; Bogenberger, Klaus · 2024 · DOAJ

DOI: 10.52825/scp.v5i.1099

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of calibrating microscopic traffic simulations to accurately reflect real-world traffic conditions, specifically focusing on vehicle counts and speed distributions. Accurate modeling is critical for traffic engineering and policy evaluation, yet traditional calibration methods using induction loops or manual counts are often costly. While the simulation software SUMO provides tools like `flowrouter` and `routesampler` to generate demand from traffic counts, these tools only perform *a priori* optimization and lack mechanisms for speed calibration, leading to non-negligible deviations from real data. To overcome these limitations, the authors propose a robust, two-step optimization method that utilizes GPS-derived vehicle count and speed measurements. The methodology consists of an *a priori* optimization phase followed by an *a posteriori* optimization phase. The first step employs Integer Linear Programming (ILP) to determine the frequency of specific routes within a generated set, aiming to match real vehicle counts on detector edges. This process formulates a minimization problem where route multipliers are adjusted to stay within the bounds of recorded vehicle counts. The second step integrates an Evolutionary Algorithm (EA) to refine the simulation parameters. Using the ILP results as an initial population, the EA optimizes both route multipliers and the maximum allowed vehicle speed for each detector edge. This *a posteriori* loop minimizes the deviation between simulated and real traffic data, specifically targeting mean speed accuracy, which the ILP step alone cannot address. The proposed method was validated as a proof of concept using a subnetwork model of Friedrichshafen, Germany, covering a three-kilometer main track. The study utilized a commercial GPS dataset from TomTom, aggregated from 36 months of data (2017–2019) into representative weekday and weekend profiles. The results were compared against SUMO’s native `flowrouter` and `routesampler` tools using Mean Absolute Error (MAE) as the goodness-of-fit metric. The ILP approach demonstrated superior performance in matching vehicle counts compared to the standard SUMO tools across all time intervals. Furthermore, the combined ILP and EA approach significantly improved the correlation with recorded mean vehicle speeds, achieving better calibration results than the ILP method alone. The EA optimization was tested on a specific high-demand time interval (6–8 a.m.), showing that the two-step process effectively reduces deviations in both count and speed metrics. The significance of this work lies in providing a generic, network-independent calibration framework that outperforms existing ready-to-use tools in SUMO. By incorporating speed calibration alongside count optimization, the method offers a more realistic reproduction of traffic dynamics. The authors conclude that this approach is scalable for large-scale traffic calibration and has potential applications in generating scenarios for Advanced Driver Assistance Systems (ADAS) and Autonomous Driving (AD) testing. Future work aims to extend the optimization to larger time ranges and incorporate additional data features to further enhance simulation fidelity.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified_with_issues.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• exposure measurement